Significance. Propensity score methods with a dichotomous treatment (exposed v. unexposed) have been extensively evaluated and proven to be an effective tool for reducing bias and balancing the distribution of confounders between comparison groups in observational studies. Propensity score methods to handle multi- level exposure, with and without ordering, and continuous exposures have been developed. However, there is very limited experience in the application of these methods in either simulated or empirical studies. Important information in establishing both the benefits and risks of medications includes the estimation of a dose effect, duration effect, and multiple medication effect in comparative effectiveness and safety investigations. Real- world medication use involves all three of these issues, and researchers need a thoroughly evaluated analytical method to apply in such scenarios. We will determine the generalizability and practical insight of the performance of propensity scores when considering these three important issues: medication dose, multiple medications, and continuous duration of therapy. This project addresses an important gap in the current analytical approach to evaluating the comparative effectiveness of medications, and will enhance and improve the quality of observational comparative effectiveness research.
Specific Aims. We will (1) Evaluate the performance of propensity scores when considering a dose-response relationship of medication exposure and the effects of multiple medication exposures;and (2) Evaluate the performance of propensity scores when considering the effects of continuously varying duration of therapy. Research Design. We will conduct both simulation studies and analyses of real data to answer comparative effectiveness research questions for both specific aims. A propensity score is defined as the conditional probability that a patient receives a particular treatment given all her/his other observed covariates. Such treatment can vary with respect to type, dose, and duration of therapy. Appropriate statistical models will be used to estimate this probability for each subject according to the nature of her/his exposure. Dose and duration effects, and difference in effect among medications will be further estimated. Performance of four common applications of propensity scores and four scenarios for which propensity scores are often applied will be evaluated. Results will be further compared with results from conventional multivariable regression models. Impact. The project will address critically important issues encountered in observational comparative effective research. It will have a wide public health impact by providing researchers with a thoroughly evaluated analytical approach to conduct such studies, and will enhance and improve the quality of observational comparative effectiveness research. .
This project will provide a critically needed and thoroughly evaluated statistical technique to conduct comparative effectiveness research studies using observational data. We will evaluate the application of propensity score methods to common situations when medication dose effect, duration effect, and multiple medication exposures need to be considered. This project addresses an important gap in the current analytical approach to medication comparative effectiveness, and will enhance and improve the quality of observational comparative effectiveness research, and thus have a wide public health impact.
|Wu, Pingsheng; Escobar, Gabriel J; Gebretsadik, Tebeb et al. (2018) Effectiveness of Respiratory Syncytial Virus Immunoprophylaxis in Reducing Bronchiolitis Hospitalizations Among High-Risk Infants. Am J Epidemiol 187:1490-1500|
|Bobo, William V; Epstein Jr, Richard A; Hayes, Rachel M et al. (2014) The effect of regulatory advisories on maternal antidepressant prescribing, 1995-2007: an interrupted time series study of 228,876 pregnancies. Arch Womens Ment Health 17:17-26|
|Hayes, Rachel M; Wu, Pingsheng; Shelton, Richard C et al. (2012) Maternal antidepressant use and adverse outcomes: a cohort study of 228,876 pregnancies. Am J Obstet Gynecol 207:49.e1-9|